Bottom line: A single malformed Base64 string can give an attacker code execution on unpatched Java runtimes, often without logging in. Fixes were released on November 17, 2026. Inventory your Java runtimes, patch identity providers first, and use a startup flag where you cannot patch today. Our full advisory has every detail. 

 What we see  

  • The flaw: On recent Intel and AMD x86-64 processors with AVX-512 VBMI, Java’s hand-tuned Base64 decoder miscalculates the output length for the padded tail of a short malformed input. The result is a write of 62 or 63 bytes past the end of its buffer, depending on the padding. Java rejects the input, but only after the write.  
  • The impact: The bytes written are fixed, so on its own the overflow usually crashes Java. Calif, the firm that found the flaw, reports shaping server memory so the overflow enlarges a neighboring array, which gives read and write access and then code execution. It demonstrated this without authentication against default installations of Keycloak, self-managed Atlassian Jira and Confluence, JFrog Artifactory, Jenkins, and PingFederate, and with a login against Apache Kafka, Solr, Elasticsearch, Gerrit, and Spinnaker. The same corruption can also expose heap memory, including private keys, session tokens, and credentials, without any code execution.  
  • Why login is often not needed: Applications decode Base64 before authentication, in HTTP Basic authentication headers, SAML artifacts, CSRF tokens, and file uploads.  
  • Side effects: Failed attempts often crash the Java process, and an attacker can often retry. Calif also reports that a Java client can be compromised by connecting over TLS to a malicious server. 
  • Exploitation status (as of November 18, 2026): We are not aware of confirmed exploitation, and the CVE is not in the CISA Known Exploited Vulnerabilities (KEV) catalog. However, the root cause is public, a triggering input has been visible since September 9, 2026, and independent teams have reproduced crashes and the memory leak in lab conditions, though not code execution as of November 18.  
  • Important Note : There is no universal exploit, because each product needs its own memory layout and timing. That slows mass exploitation but does not protect a specific, high-value target such as an identity provider.  

Are you affected? 

A system is vulnerable only when all three conditions are true: an unpatched runtime, a specific processor, and attacker-controlled Base64 that reaches Java’s standard decoder. 

Condition   Affected   Not affected  
Java runtime   Unpatched OpenJDK-based builds from 17.0.6 onward on Java 17, and all Java 21, 25, and 27 builds before the fix, from any vendor. Oracle also lists its Java 8 performance build.   Java 17 releases before 17.0.6, standard Java 8 and Java 11, and fixed builds (Oracle: 17.0.21.1, 21.0.13.1, 25.0.5.1, 27.0.1.1)  
Processor   x86-64 with AVX-512 VBMI, including Intel Xeon Ice Lake through Granite Rapids and AMD EPYC and Ryzen Zen 4 and newer. The CPU must expose avx512f, avx512bw, avx512vbmi, and bmi2   ARM64, and x86-64 processors without AVX-512 VBMI  
Reachability   Attacker-controlled Base64 reaches Java’s standard decoder (login headers, tokens, uploads)   No path for external input to reach the standard Base64 decoder  

 How to check 

  1. Runtime: Build an inventory of the runtime each application actually uses. Bundled runtimes do not get OS updates, so check the one each application ships with. 
  2. Processor: Run the Linux command in the advisory. A host is affected only if avx512f, avx512bw, avx512vbmi, and bmi2 all appear in its CPU flags (/proc/cpuinfo). Use the result to prioritize, not to dismiss a finding, because a cloud resize can silently move a workload onto an affected processor. 
  3. Reachability: This is the hardest to settle from the inside, because it depends on how each application is deployed and what sits in front of it. 

What else is in scope 

  • Self-managed Jira and Confluence: The risk applies to self-managed Data Center deployments. Atlassian manages the Java runtime for Atlassian Cloud. Atlassian installers typically ship their own bundled runtime, so operating system patches alone do not fix a self-managed instance. 
  • Spring Boot: Treat every Spring Boot service on an affected runtime as in scope. Researchers reached code execution in sample Spring Boot applications, where login handling, tokens, and file uploads often decode Base64 before a user is authenticated. 
  • Apache Tomcat: It was not tested and no advisory links it to this CVE. It runs a large share of Java web applications and is embedded in Spring Boot by default, so include every Tomcat server in your inventory. 
  • Anything else: If you cannot rule out that attacker input reaches Java’s built-in Base64 decoder, assume it does. Vendor advisories cover only Java 17, 21, 25, and 27, so move any other release to a fixed build rather than assuming it is safe. 

What to do: patch first 

Updating the Java runtime is the only complete fix. 

Fix availability (as of November 18, 2026) 

  • Fixed builds shipped: Oracle, Amazon (Corretto), Azul (Zulu). For Oracle, only builds ending in .1 are patched. 
  • Not yet released: Red Hat and Eclipse Temurin 
  • Status unconfirmed: Microsoft Build of OpenJDK, BellSoft Liberica. Check each vendor’s page. 
  • Bundled runtimes: Products that ship their own Java runtime need an update from the product vendor. 

Patch in this order 

  1. Identity providers: Keycloak and PingFederate 
  2. Jira, Confluence, Jenkins, and Artifactory (for Jira and Confluence, apply Atlassian’s update or the startup flag) 
  3. Other internet-facing Java services that handle logins, tokens, or uploads  
  4. Internal systems and Java clients 

If you cannot patch today 

  1. Add -XX:-UseBASE64Intrinsics to the Java startup options and restart the service. This turns off only the accelerated Base64 decoder. Applications that decode large volumes of Base64 may slow down, so test first. 
  2. Alternatively, use -XX:UseAVX=2. It turns off AVX-512 entirely, at a broader performance cost. Java on ARM64 rejects this option and will not start. 
  3. Remove either option after you have patched. 

Avoid: -XX:DisableIntrinsic=_base64_decodeBlock on its own. It can stop Java from starting. 

Verify: A patch or a flag takes effect only after Java restarts. Confirm on the running process: 

jcmd <pid> VM.version  

jcmd <pid> VM.flags  

VM.version should show a fixed build. For Oracle that is a version ending in .1, such as 17.0.21.1, and other vendors use their own version strings. If you set the startup flag, VM.flags should list -XX:-UseBASE64Intrinsics. 

Elasticsearch 

Elasticsearch ships with a bundled Java runtime, and Calif demonstrated code execution against it after login. Some attacks depend on a request arriving in pieces, so a reverse proxy that waits for complete requests (nginx does by default) blocks those. This attack works in a single complete request, so request buffering does not stop it. Treat clusters with anonymous access or weak credentials as exposed without a login. Apply Elastic’s update, or add -XX:-UseBASE64Intrinsics in a new .options file under config/jvm.options.d/ (/etc/elasticsearch/jvm.options.d/ for DEB and RPM installs) and restart each node. Do not edit the root jvm.options file. 

How to spot attempts 

Failed attempts often crash Java, so watch for unexpected service restarts and new hs_err_pid<pid>.log files, which the JVM writes on a fatal crash. Crashes during garbage collection are also a known sign of this corruption. Then list every place where attacker-controlled Base64 reaches Java before login: HTTP Basic headers, SAML artifacts, CSRF tokens, and file uploads. 

Find the Java services you expose to the internet 

Check Point Exposure Management can identify internet-facing assets that appear to run Java, including the version, so you know where to start. This is a passive check of technologies and versions. It does not exploit anything or send attack traffic to your services. 

A technology check shows where Java is present. It cannot confirm the processor or whether attacker-controlled input reaches Java’s Base64 decoder, so use the results to prioritize and still run the runtime and processor checks above. 

Beyond this flaw, Exposure Management helps you find, prioritize, validate, and fix the exposures that matter most across your environment: 

  • Threat Intelligence: Know which threats and vulnerabilities attackers are targeting. 
  • CAASM: Get a continuously updated view of your assets and where they are exposed. 
  • Vulnerability Prioritization: Focus on the exposures with real risk, not the longest list. 
  • Agentic Exposure Validation (AEV): Test whether an exposure is exploitable. 
  • Safe Remediation: Fix confirmed exposures without disrupting the business. 

Contact Us 

You may also like